On Thu, Feb 11, 2016 at 07:45:11AM +0100, Andrzej Hajda wrote:

On 02/11/2016 04:38 AM, Guenter Roeck wrote:

quoted

On 02/10/2016 07:21 AM, Arnd Bergmann wrote:

quoted

On Tuesday 09 February 2016 18:37:46 Guenter Roeck wrote:

quoted

On 02/09/2016 07:26 AM, Arnd Bergmann wrote:

quoted

On Tuesday 09 February 2016 07:08:59 Guenter Roeck wrote:

quoted

IS_ERR_VALUE() assumes that its parameter is an unsigned long.
It can not be used to check if an unsigned int reflects an error.
Doing so can result in the following build warning.

drivers/tty/serial/digicolor-usart.c: In function ‘digicolor_uart_probe’:
include/linux/err.h:21:38: warning:
          comparison is always false due to limited range of data type
drivers/tty/serial/digicolor-usart.c:485:6: note:
          in expansion of macro ‘IS_ERR_VALUE’

If that warning is seen, an error return from platform_get_irq() is missed.

The patch looks correct to me, but what compiler version and which kernel
tree is it that triggered the warning?

Andrzej Hajda just modified the definition of IS_ERR_VALUE(), and the
changes are still under discussion, but I don't see that warning with
any of the versions.

I see it with gcc 5.1 and 5.2 (and W=1). I did not see / notice Andrzej's patch.

I agree that fixing the problem in IS_ERR_VALUE() is preferrable.

Ah, W=1 explains it. We are still debating about the proper solution. Al Viro
pointed out that most users of IS_ERR_VALUE() shouldn't be using it at all,
so your patch is probably best here after all.

Yes, after looking into it some more I agree. Coccinelle should be able to handle
most of the conversions automatically. I actually tried to write a script, just
for the fun of it, but it misses some of the problem cases in patch mode.
Maybe I get it working tonight.

Guenter

Maybe it can be helpful. The script which classifies all IS_ERR_VALUE
usages as
correct, signed, incorrect and unknown (cocci was not able to detect type).
Adding patch mode should be quite simple.

// Options: --all-includes --include
include/uapi/linux/netfilter_ipv6/ip6_tables.h --include
include/uapi/linux/netfilter_ipv4/ip_tables.h --include
include/uapi/linux/netfilter_arp/arp_tables.h --include
include/uapi/linux/netfilter/x_tables.h --include include/linux/mm_types.h

virtual context
virtual report

@r@
typedef bool, u8, u16, u32, u64, s8, s16, s32, s64, __u64, dma_addr_t,
Elf32_Addr, Elf64_Addr;
{unsigned long, size_t} ev;
{short, int, long, s16, s32, ssize_t} ew;
{char, unsigned char, unsigned short, unsigned int, long long, unsigned
long long, bool, u8, u16, u32, u64, s64, __u64, dma_addr_t, Elf32_Addr,
Elf64_Addr} ei;
expression eu;
position p;
@@

IS_ERR_VALUE(\(ev \| ew \| ei \| eu \))@p

@script:python depends on report@
p << r.p = [];
ev << r.ev = "";
ew << r.ew = "";
ei << r.ei = "";
eu << r.eu = "";

This doesn't work for me; it bails out at the assignments. If I remove
the assignments, it does not report anything.

Any idea what might be wrong ?

Thanks,
Guenter

@@

t = ""
if ev: t = "correct"
if ew: t = "signed"
if ei: t = "incorrect"
if eu: t = "unknown"


e = ev + ew + ei + eu;
if t:
    msg = "WARNING: %s argument type in IS_ERR_VALUE(%s)" % (t, e)
    coccilib.report.print_report(p[0], msg)